We have a decklink 8k card,and We want to use cuda to process the captured frame. We found there is technoloy called GPUDirect.However after we read the related material and example of decklink about gpudirect.It seem that it should have a shared pinned memory ,then the data transfer is that :decklink card -> shared pinned memory -> video memory. Actually it still consume time for us.So we want to know whether the data transfer can like that :decklink card ->video memory. there exits other two gpuDirect methods :P2P and GPUdirect RDMA.We read a lot about them ,but still does’t make sense how to do it.
So we want to ask is it possible that the decklink card can transfer the data directly to the video memory of nvdia card using P2P or GPUdirect RDMA? If it is possible ,can somebody give us some tips or tutorials .
Thank you